arXiv:2604.20909v1 Announce Type: new Abstract: Downhole drilling telemetry presents a fundamental labeling asymmetry: surface sensor data are generated continuously at 1~Hz, while labeled downhole measurements are costly, intermittent, and scarce. Current machine learning approaches for downhole metric prediction universally adopt fully supervised training from scratch, which is poorly suited to
Do Masked Autoencoders Improve Downhole Prediction? An Empirical Study on Real Well Drilling Data
Aleksander Berezowski, Hassan Hassanzadeh, Gouri Ginde·arXiv cs.LG··1 min read
a
Continue reading on arXiv cs.LG
This article was sourced from arXiv cs.LG's RSS feed. Visit the original for the complete story.